From circuit topology to behavioural model

of power amplifier
dedicated to radar applications

F.X. Estagerie, T. Reveyrand, S. Mons, R. Quéré, L. Constancias and P. Le Helleye

Published in
Electronics Letters -- 12 April 2007 -- Volume 43, Issue 8, p. 477-479

© 2007 IET Personal use of this material is permitted. However, permission to

reprint/republish or redistribute this material for advertising or promotional purposes or
for creating new collective works for resale or redistribution to servers or lists, or to reuse
any copyrighted component of this work in other works must be obtained from the IET.
From circuit topology to behavioural model shows an example of the model’s topology. In the following, the Letter
of power amplifier dedicated to radar focuses on the modelling of transistors.
applications
F.X. Estagerie, T. Reveyrand, S. Mons, R. Quéré,
L. Constancias and P. Le Helleye

A new behavioural model of a power amplifier (PA) at system level is

presented, dedicated to radar applications. This model, which is able to
take into account output loading impedance mismatch, is based on a
similar topology to the PA’s circuit. The model, implemented in
Agilent Advanced Design System (ADS), is validated for different
loading impedances, until VSWR ¼ 2 (voltage standing wave ratio),
on the PA’s bandwidth. This approach permits an extraction, resulting
from simple CW measurements or simulations, and easy model
implementation. This work has been supported by DGA (French
Defense and Security Agency).

Introduction: In recent years the development of active electronically

scanned array (AESA) radar has been significant. This technology
will equip most of next generation military radar systems [1]. In the
framework of the development of such systems, an accurate ‘system’
simulation is needed. Indeed, nowadays, cost reduction of such
systems is a key issue for the European defence industry (an AESA
is two times more expensive than a passive electronically scanned
array) and that implies the use of accurate simulation tools to decrease
margins taken on component specifications. Such simulations must be
able to optimise the design of the whole radar functional chain by
enabling design engineers to predict and analyse the impact of
microwave components on ‘system’ performance. Unfortunately,
design models of microwave components cannot be used for such a
simulation because it would lead to huge computing time. That is why
technical solutions have been explored to reduce complexity of
simulation accuracy with radar signals. Today, simulation tools
dedicated to AESA radar describe RF systems in only a unilateral
Fig. 1 Grey Box approach: assembly of linear and nonlinear sub-models,
based on same topology of design circuit
Modelling of nonlinear parts: Transistors are modelised thanks to
nonlinear scattering functions. If we make the assumption that a
transistor is considered to be a nonlinear two-port circuit at funda-
mental frequency, without memory effects, it is defined by the
following relationship:
b~ i ¼ fNLi ð~a1 ; a~ 1* ; a~ 2 ; a~ *2 Þ ð1Þ
where i ¼ 1, 2, b̃ 1, b̃ 2 and ã 1, ã 2 are, respectively, the reflected and
incident power waves at the two ports. In conditions of weak loading
impedance mismatch, ã 2 can be considered weak compared to ã 1;
moreover, if ã 1 is considered like the reference wave, Taylor series
expansions limited to the first order lead to write [3]:
way and with insufficient accuracy. Nevertheless, design constraints !  
 
!  
of AESA lead to significant load mismatch (up to VSWR ¼ 2) with b~ 1 S11 a~ 1  S12 a~ 1  a~ 1
¼  
 


varying phase in microwave chains implying the need for a nonlinear b~ 2 S21 a~ 1  S22 a~ 1  a~ 2
bilateral model. It appears that development of a powerful simulation  
!
D   *
!
0 S12 a~ 1 a~ 1
tool requires an accurate model of the power amplifier (PA) in order to þ  

ð2Þ
D  
quantify its impact on transmission=reception (T=R) modules and 0 S22 a~ 1 a~ * 2
then on emitted signal characteristics. Indeed, the PA undergoes many Sij(jã 1j) represent nonlinear scattering functions, as a function of the
disturbances in AESA radars, particularly output loading impedance incident wave’s magnitude. To extend the capabilities of prediction to
mismatch. In fact, impedance presented to the output port of the PA the PA’s bandwidth, the frequency dispersion of the transistor must be
varies because of AESA controls. The difference between this considered. An efficient solution consists of parameterising Sij(jã 1j)
impedance and PA load required in the case of optimal match with appropriate coefficients. Thus (2) becomes:
condition can be huge. PA behavioural models employed in these !  
 
!  
simulations, simple gain or AM=AM-AM=PM, are insufficient. b~ 1 S11 a~ 1 
c11 ðOÞ S12 a~ 1 
c12 ðOÞ a~ 1
Recently, more efficient behavioural models were developed [2–5] ¼  
 


~b2 S21 a~ 1 
c21 ðOÞ S22 a~ 1 
c22 ðOÞ a~ 2
but their implementation and their extraction are complex. Therefore,
a~ *1
 
D ! !
D  
we present the principle of a novel approach, called ‘Grey Box’, and 0 S12 a~ 1
c12 ðOÞ
its results. This approach is particularly useful when the topology of þ  
D
the PA is known at the design level. In order to satisfy AESA radar
needs, this model must be compliant with the following specifications:
it should be accurate with VSWR up to 2; it should take into account
thermal effects; and it should take into account fast and slow memory
effects.
Model structure: The Grey Box model is directly derived from the
topology of the amplifier. Thus, this model is divided into linear and
nonlinear sub-models, respectively associated to passive and active
elements. To take into account output impedance mismatch, linear and
nonlinear models are described in a bilateral way. Passive elements are
described thanks to their linear scattering parameters. Nonlinear
scattering functions [3], which are an extension of linear scattering
parameters applied to the nonlinear case, permit us to describe the
transistors. If the PA has several transistors by the amplification stage,
the model topology can be simplified since the impedances presented
to active cells at each stage are very close. As a result, passive cells
can be reduced to a two-port structure. Therefore, a single nonlinear
sub-model is necessary for each stage if input=output currents are
divided by the number of transistors for this stage. This model
simplification enables us to use a single extraction for all transistors
of a same power amplifier stage and to enhance time cost. Fig. 1
ELECTRONICS LETTERS 12th April 2007 Vol. 43 No. 8
D  

ð3Þ
0 S22 a~ 1
c22 ðOÞ a~ *2
cij(O) is frequency function, normalised to one, where O ¼ o o0. The
identification of nonlinear scattering functions and coefficients are
obvious because it requires a simple extraction. In order to solve (2)
nonlinear scattering functions are extracted, at the PA’s operating
frequency, from three different loading impedance CW measurements,
with a vectorial network analyser (VNA) [6], or CW harmonic balance
simulations. Coefficients are extracted, for low level signal excitation,
thanks to the same method as nonlinear scattering functions for several
frequency points on the PA’s bandwidth.
Model implementation: The Grey Box model has been implemented
in Agilent ADS, likewise in [5]. Data files from S-parameters passive
element’s simulation describe linear parts. Frequency device domain
(FDD) is used in order to modelise transistors, where (3) defines the
relationships between the input=output ports.
Results: To validate the proposed modelling approach, it was applied
to a PA design project in Agilent ADS. It is an X-band, 18 dB gain
amplifier, which contains four GaAs HBTs in the driver’s stage and
eight in power’s stage. Fig. 2 shows an example of comparison
between circuit-level simulation and the Grey Box model at operating
frequency, the PA was terminated with different loads far beyond its
optimum load (VSWR ¼ 2). At different input levels, Fig. 3 shows a
comparison of output power response of the circuit level and
behavioural model against the PA’s bandwidth.
Fig. 2 Simulation-based Grey Box model (solid lines) compared to circuit
(circles, cross, rhombus triangles)
Gain in decibels and AM-PM in degrees against input power in dBm for the two-
stage amplifier of Fig. 1 terminated with different loads at PA’s operating frequency
Fig. 3 Simulation-based Grey Box model (solid lines) compared to circuit
(circles)
At different input levels, output power in dBm against frequency offset from
carrier in MHz for PA’s nominal load
ELECTRONICS LETTERS 12th April 2007 Vol. 43 No. 8
Conclusion: We have presented a novel behavioural model based on
an actual PA design. It is divided into several simple sub-models
corresponding to linear and nonlinear parts of this PA. The tests,
carried out on a two-stage PA, have been validated in frequency
domain. This new approach has several advantages: it enables a good
prediction of the PA’s behaviour in the case of strong output loading
impedances mismatch (until VSWR ¼ 2), it is extracted from simple
CW measurements or simulations, and its implementation in a
simulator is obvious. This model should be used in the time domain
environment in order to evaluate the PA’s behaviour submitted to pulse
signal excitations for radar applications and finally it should be
integrated in the DGA=CELAR AESA radar simulation tool.
Acknowledgment: This work has been supported by the French
Defence and Security Agency (DGA).
# The Institution of Engineering and Technology 2007
2 October 2006
Electronics Letters online no: 20073059
doi: 10.1049/el:20073059
F.X. Estagerie, T. Reveyrand, S. Mons and R. Quéré (XLIM –
University of Limoges, 123 avenue Albert Thomas, Limoges Cedex
87060, France)
E-mail: estagerie@xlim.fr
L. Constancias and P. Le Helleye (DGA=CELAR–BP 7419 – 35174
Bruz Cedex, France)
References
1 Hommel, H., and Feldle, H.P.: ‘Current status of airborne active phased
array (AESA) radar system and future trends’. 34th European Microwave
Conf., Amsterdam, The Netherlands, 2004
2 Ngoya, E., and Soury, A.: ‘Modeling memory effects in nonlinear
subsystems by dynamic Volterra series’. IEEE MTT-S Int. Microwave
Symp., June 2003
3 Verspecht, J.: ‘Scattering functions for nonlinear behavioral modeling in
the frequency domain’. Fundamentals of Nonlinear Behavioral
Modeling: Foundations and Applications Workshop, IEEE MTT-S Int.
Microwave Symp., June 2003
4 Soury, A., Ngoya, E., and Rousset, J.: ‘Behavioral modeling of RF and
microwave circuit blocs for hierarchical simulation of modern
transceivers’. IEEE MTT-S Int. Microwave Symp., June 2005
5 Root, D.E., Verspecht, J., Sharrit, D., Wood, J., and Cognata, A.: ‘Broad-
band poly-harmonic distortion (PHD) behavioral models from fast
automated simulations and large-signal vectorial network
measurements’, IEEE Trans. Microw. Theory Tech., 2005, 53, (11),
pp. 3656–3664
6 Reveyrand, T., et al.: ‘A time domain envelope vectorial network analyser
for non-linear measurement based modeling accounting impedance
mismatches’. IMTC 2006, Sorrento, Italy, April 2006